QoS hot changes changes (tc)

QoS hot changes changes (tc)

[Raúl Hernández]

Hi there,

I am trying to characterize the situation in which we perform a change
in the shaping done by the Linux kernel (tc+qdisc), specifically when
modifying the bandwidth as a hot change while shaping with a previous
bw value. The system I am describing performs shaping per stream
(uplink and/or dowinlink or total), just marking the TOS of the ip
packet and provisioning the kernel via 'tc' to shape the packets.

My question is whether someone has tried to modify bw parameters, ie:
move from 90 to 45 kb/s while the queues are already being populated
with trafffic .Do not know if the shaper algorithm is able to adjust
the bw without service disruption/experiment transitory behavior as I
guess the algorithm must have some kind of feedback which depends of
the bytes already shaped and the value of the bw itself (ie: leaky
bucket).

Any experience with this kind of changes :-) ? Thx !

Raul

-- 
"We cannot change the cards we are dealt, just how we play the hand".

Re: QoS hot changes changes (tc)

[Jarek Poplawski]

Raúl Hernández wrote, On 06/26/2009 03:19 PM:

> Hi there,
> 
> I am trying to characterize the situation in which we perform a change
> in the shaping done by the Linux kernel (tc+qdisc), specifically when
> modifying the bandwidth as a hot change while shaping with a previous
> bw value. The system I am describing performs shaping per stream
> (uplink and/or dowinlink or total), just marking the TOS of the ip
> packet and provisioning the kernel via 'tc' to shape the packets.
> 
> My question is whether someone has tried to modify bw parameters, ie:
> move from 90 to 45 kb/s while the queues are already being populated
> with trafffic .Do not know if the shaper algorithm is able to adjust
> the bw without service disruption/experiment transitory behavior as I
> guess the algorithm must have some kind of feedback which depends of
> the bytes already shaped and the value of the bw itself (ie: leaky
> bucket).
> 
> Any experience with this kind of changes :-) ? Thx !

No experience in measuring the effects of such changes, but looking
into code shows it depends on a shaper algorithm: tbf seems to reset
the queue, htb doesn't.

Jarek P.

Re: QoS hot changes changes (tc)

[Raúl Hernández]

Thanks Jarek ! I will try to test them running traffic to see how they
are doing. I miss some documentation related to these stuff :-(.

Regards,
Raul

2009/6/27 Jarek Poplawski <jarkao2@gmail.com>:
> Raúl Hernández wrote, On 06/26/2009 03:19 PM:
>
>> Hi there,
>>
>> I am trying to characterize the situation in which we perform a change
>> in the shaping done by the Linux kernel (tc+qdisc), specifically when
>> modifying the bandwidth as a hot change while shaping with a previous
>> bw value. The system I am describing performs shaping per stream
>> (uplink and/or dowinlink or total), just marking the TOS of the ip
>> packet and provisioning the kernel via 'tc' to shape the packets.
>>
>> My question is whether someone has tried to modify bw parameters, ie:
>> move from 90 to 45 kb/s while the queues are already being populated
>> with trafffic .Do not know if the shaper algorithm is able to adjust
>> the bw without service disruption/experiment transitory behavior as I
>> guess the algorithm must have some kind of feedback which depends of
>> the bytes already shaped and the value of the bw itself (ie: leaky
>> bucket).
>>
>> Any experience with this kind of changes :-) ? Thx !
>
> No experience in measuring the effects of such changes, but looking
> into code shows it depends on a shaper algorithm: tbf seems to reset
> the queue, htb doesn't.
>
> Jarek P.
>



-- 
"We cannot change the cards we are dealt, just how we play the hand.